variation op• 数据种类:table × video → table
• 调用: import fullseye as fs; fs.ledger.varmodel_calibrate(model, good_images, far=None)(要直接调用实现,import varmodel; varmodel.varmodel_calibrate(model, good_images, far=None);从台账取用则 opsspc.get("varmodel_calibrate"))
用未参与拟合的良品重新求容许带半宽 k(用于不满足正态假设的真实图像,table)。
> 以下的详细说明为原文 —— 摘要与标题已翻译。
model is a varmodel_fit table; good_images (m, H, W) are good samples that were NOT used to fit it. Each is scored (varmodel_score, align="translate") and the pooled |z| of all valid pixels gives the empirical (1 - far) quantile, which replaces k (far defaults to the model's declared rate). Defects are never looked at: the threshold still comes from the good distribution, only measured instead of assumed. Real images break the per-pixel independent-normal model (residual misregistration, compression, lighting drift); the closed-form k then under-states the false alarms and this is the honest fix.
Returns a copy of the model with k / lower / upper replaced and k_declared (the closed-form k), k_source = held-out quantile, calibration_n_images, calibration_n_pixels and calibration_far_declared_k (the false-alarm rate the closed-form k actually had on these good images).
Raises ValueError on an empty / malformed / non-finite set, a shape mismatch, a model without the fitted keys, far outside its range, or fewer than 10/far calibration pixels (the quantile would rest on fewer than 10 exceedances).
• 示例数据目录(下载 URL / 许可证) —— 2-D 用 skimage.data(BSD/公有领域)加合成图,3-D 给出真实数据源(Stanford/PDS 等)的下载 URL。
• 算子来历与参考文献 —— 该算子族所依据的研究/方法出处。
• 算法的正典(作者、年份)与用途见上面的族使用指南。
• poc_good_sample_variation_model — py -3.11 examples/poc_good_sample_variation_model.py
table 作为输入)msa_anova_table · msa_gauge_rr · msa_bias_linearity · msa_attribute_agreement · gum_standard_uncertainty · gum_propagate · gum_expanded · gum_monte_carlo
variation)varmodel_alignment · varmodel_band_k · varmodel_fit · varmodel_score
*Provenance: varmodel.py — SPC 算子登记表。本条目由 tools/opdocs.py md 自动生成(请勿手工编辑)。*
© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.